Object recognition apparatus and method

1. An object recognition apparatus comprising:

a learning unit configured to:

determine an image feature vector of a first image by applying, to the first image, a convolution network for extracting features from a plurality of image learning sets including the first image and a sample segmentation map of the first image;

determine a first segmentation map of the first image by applying a deconvolution network to the determined image feature vector; and

determine a weight of the convolution network and a weight of the deconvolution network based on the sample segmentation map and the first segmentation map; and

a segmentation map determiner configured to determine a second segmentation map of a second image through the convolution network using the determined weight of the convolution network and through the deconvolution network using the determined weight of the deconvolution network.

2. The object recognition apparatus of claim 1 , wherein the convolution network comprises:

a plurality of convolution layers configured to generate at least one feature map by performing convolution; and

a pooling layer configured to down-sample the at least one feature map between the plurality of convolution layers; and

wherein the deconvolution network comprises:

a plurality of deconvolution layers configured to generate at least one intermediate segmentation map by performing deconvolution; and

an unpooling layer configured to up-sample the at least one intermediate segmentation map between the plurality of deconvolution layers.

3. The object recognition apparatus of claim 2 , wherein the weight of the convolution network is a coefficient of a convolution mask for the plurality of convolution layers, and

the weight of the deconvolution network is a coefficient of a deconvolution mask for the plurality of deconvolution layers.

4. The object recognition apparatus of claim 2 , wherein the pooling layer is a max-pooling layer.

5. The object recognition apparatus of claim 4 , wherein the unpooling layer corresponds to the max-pooling layer.

6. The object recognition apparatus of claim 1 , wherein the plurality of image learning sets comprise a third image, which is obtained by extracting only one object from the first image, and a second sample segmentation map of the third extracted image.

7. An object recognition method comprising:

determining an image feature vector of a first image by applying, to the first image, a convolution network for extracting features from a plurality of image learning sets including the first image and a sample segmentation map of the first image;

determining a first segmentation map of the first image by applying a deconvolution network to the determined image feature vector;

determining a weight of the convolution network and a weight of the deconvolution network based on the sample segmentation map and the first segmentation map; and

determining a second segmentation map of a second input image through the convolution network using the determined weight of the convolution network and through the deconvolution network using the determined weight of the deconvolution network.

8. The object recognition method of claim 7 , wherein the convolution network comprises:

a plurality of convolution layers configured to generate at least one feature map by performing convolution; and

a pooling layer configured to down-sample the at least one feature map between the plurality of convolution layers; and

wherein the deconvolution network comprises:

a plurality of deconvolution layers configured to generate at least one intermediate segmentation map by performing deconvolution, and

an unpooling layer configured to up-sample the at least one intermediate segmentation map between the plurality of deconvolution layers.

9. The object recognition method of claim 8 , wherein the weight of the convolution network is a coefficient of a convolution mask for the plurality of convolution layers, and

the weight of the deconvolution network is a coefficient of a deconvolution mask for the plurality of deconvolution layers.

10. The object recognition method of claim 8 , wherein the pooling layer is a max-pooling layer.

11. The object recognition method of claim 10 , wherein the unpooling layer corresponds to the max-pooling layer.

12. The object recognition method of claim 7 , wherein the plurality of image learning sets comprise a third image, which is obtained by extracting only one object from the first image, and a second sample segmentation map of the third extracted image.

13. An image magnification apparatus comprising:

a learning unit configured to:

determine an image feature vector of a first image by applying, to the first image, a convolution network for extracting features from a plurality of image learning sets including the first image and a sample super-resolution image of the first image;

determine a first super-resolution image of the first image by applying a deconvolution network to the determined image feature vector; and

determine a weight of the convolution network and a weight of the deconvolution network based on the sample super-resolution image and the first super-resolution image; and

a super-resolution image determiner configured to determine a second super-resolution image of a second image through the convolution network using the determined weight of the convolution network and through the deconvolution network using the determined weight of the deconvolution network.